Hi Jed, Thanks for reporting this. I don't know the exact history of the configure tests, but my "fix" was to make the results consistent across our testing platforms. From your email, it appears there is larger issue we need to address in order to do what's intended. I'll create a new ticket. Ken On 01/08/2014 02:19 PM, "Antonio J. Peña" wrote:

Let me discuss this with Ken. We'll get back to you.

Antonio

On 01/06/2014 06:19 PM, Jed Brown wrote:

...

Jed Brown <[email protected]> writes:

...

Pavan Balaji <[email protected]> writes:

...

This attribute is tested in configure before it’s added to mpi.h.in. Yes, the configure test is incorrect and the resulting usage in MPICH is broken (does not do what you intend). As far as I know,

#pragma weak MPI_Xyz = PMPI_Xyz

is equivalent to what was intended with the attribute syntax

__attribute__((weak, alias("PMPI_Xyz")))

(which belongs in an implementation file, not in the public header). Is there a reason to bother testing for the attribute syntax?